PoseFusion: PoseFusion: Robust Object-in-Hand Pose Estimation with SelectRNN

Project website:https://elevenjiang1.github.io/ObjectInHand-Dataset/

1. Data Extraction

We upload all our dataset in GoogleDrive, all the data are saved in rosbag format. And example code for data extraction can be available from DataExtraction Folder.

1.1 Usage

(1) Build all package

git clone git@github.com:elevenjiang1/ObjectInHand-Dataaset.git
cd ObjectInHand-Dataset/DataExtraction
catkin_make

(2) Run rosbag and see data

# Always need to publish static transform
roscd data_extraction/scripts/files
rosbag play -l tf_static.bag

# Rosbag play data
cd /file/to/Download/cracker
rosbag play rosbag play 2022-08-14-16-00-16-processed.bag

(3) Record data

# Data will saved in 
roscd data_extraction/scripts/
python extract_data.py

2. SelectLSTM

Please download the dataset first, and change all dataset path in train_network.py and test_network.py Default whole dataset path is in: /home/media/WholeDataset

Both tactile, vision, merge and selectLSTM training function is in train_network.py To train the SelectLSTM, please generate select_dataset by function generate_select_data() in test_network.py, base on yourself training network